Artificial intelligence
in personalized medicine
Enrico Glaab
Overview
(1) Introduction: Artificial Intelligence vs. Classical Statistics
(2) AI success stories in recent years
(3) AI in personalized medicine:
- Biomarker discovery
- Drug discovery
- Digital health monitoring
(4) Common pitfalls and challenges
AI vs. Statistics
Artificial Intelligence
Machine Learning
Deep Learning
Statistics
Enabling machines to think like humans
Training machines to learn a task
without explicit programming
ML using multi-layered
networks without manual
AI success stories: Image data (1)
• Image classification: unprecedented accuracies using deep learning
(e.g. “AlexNet” approach by Krizevsky et al., 2012)
AI success stories: Sequential data (2)
• Speech recognition: 30% less errors with new neural networks
(Huang et al., ACM Communications, 2012)
• Novel techniques: Recurrent Neural Networks (RNNs):
AI successes in personalized medicine - Biomarkers
Name
Test approval
(FDA-cleared and/or LDT)
Purpose
References
MammaPrint
FDA-cleared, LDT
breast cancer
risk-of-recurrence
assessment
Van’t Veer et al.,
Nature, 2002
AlloMap Heart
FDA-cleared, LDT
identifying heart
transplant recipients
with risk of cellular
rejection
Yamani et al., J Heart
Lung Transplant, 2007
Prosigna Assay /
PAM50
FDA-cleared, LDT
breast cancer risk of
distant recurrence
prediction
Nielsen et al., BMC
Cancer, 2014
Oncotype DX
LDT
breast cancer
risk-of-recurrence
assessment
Kelley et al., Cancer,
2010
Decipher
LDT
prostate cancer
metastatic risk
prediction
Marrone et al., PLoS
Curr., 2015
AI in biomedicine: Covid-19 hospital mortality prediction
• Covid-19: alterations in
common clinical blood
tests when diagnosed
• Apply ML à 90% accurate
predictions of mortality on
test set of 110 patients
(10 days in advance)
Deep learning for biomedical research: AlphaFold 2
• DeepMind’s AlphaFold 2 à major advance in protein structure prediction
• Scores > 90 in global distance test (GDT) in the CASP competition
AI in drug development
AI in drug development
• Question: Which drug-like compounds bind to a particular target protein?
• Example:
Active compounds
Inactive / low activity compounds
*
Application to SARS-CoV-2 compound screening
AI-based screening: Finding new inhibitors of viral proteins
3CLpro
and
RdRp
AI for personalized health monitoring / digital biomarkers
Digital biomarkers:
- mobile phone gyrometer & accelerometer
- gait sensors (eGaIT system)
Recommendations from literature
Data pre-processing, filtering & normalization:
à
use
cross-validation
to check if pre-processing leads to information loss
à
compare or combine
multiple pre-processing approaches
Integration of prior knowledge & multi-omics analyses:
à
Assess cost/benefit
of multi-omics analyses using prior data, or conduct pilot analyses
à
use
existing software & frameworks
for integrative biological data analysis
Ensuring model interpretability & biological plausibility:
à
use dedicated methods to build
interpretable models
(e.g. rule learning methods)
Outlook: New AI strategies to address the challenges
Increase statistical power
à
AI-based integration of prior knowledge & multi-omics data
à
Algorithmic sample matching & selection, AI for batch adjustment
Increase model robustness
à
AI-based integration of models across human & animal model data,
different experimental platforms and cohorts (Transfer learning)
Increase model interpretability
à
Structured and graph-based machine learning
Acknowledgements
Biomedical Data Science Group
New members:
Thank you for your attention!
References
1. E. Glaab, Building a virtual ligand screening pipeline using free software: a survey, Briefings in Bioinformatics (2015), 17(2), pp. 352
2. E. Glaab, Using prior knowledge from cellular pathways and molecular networks for diagnostic specimen classification, Briefings in Bioinformatics (2015), 17(3), 440 3. N. Vlassis, E. Glaab, GenePEN: analysis of network activity alterations in complex diseases via the pairwise elastic net, Statistical Applications in Genetics and
Molecular Biology (2015), 14(2), 221
4. E. Glaab, J. M. Garibaldi, N. Krasnogor. Learning pathway-based decision rules to classify microarray cancer samples, German Conference on Bioinformatics 2010, Lecture Notes in Informatics (LNI), 173, 123-134
5. E. Glaab, J. Bacardit, J. M. Garibaldi, N. Krasnogor, Using rule-based machine learning for candidate disease gene prioritization and sample classification of cancer
gene expression data, PLoS ONE, 7(7):e39932, 2012
6. E. Glaab, A. Baudot, N. Krasnogor, A. Valencia. Extending pathways and processes using molecular interaction networks to analyse cancer genome data, BMC Bioinformatics, 11(1):597, 2010
7. E. Glaab, A. Baudot, N. Krasnogor, R. Schneider, A. Valencia. EnrichNet: network-based gene set enrichment analysis, Bioinformatics, 28(18):i451-i457, 2012 8. Maes, M., Nowak, G., Caso, J. R., Leza, J. C., Song, C., Kubera, M., .et al. (2016). Toward omics-based, systems biomedicine, and path and drug discovery
methodologies for depression-inflammation research. Molecular neurobiology, 53(5), 2927-2935.
9. E. Glaab, J.P. Trezzi, A. Greuel, C. Jäger, Z. Hodak, A. Drzezga, L. Timmermann, M. Tittgemeyer, N. J. Diederich, C. Eggers, Integrative analysis of blood metabolomics and PET brain neuroimaging data for Parkinson's disease, Neurobiology of Disease (2019), Vol. 124, No. 1, pp. 555
10. E. Glaab, R. Schneider, Comparative pathway and network analysis of brain transcriptome changes during adult aging and in Parkinson's disease, Neurobiology of Disease (2015), 74, 1-13
11. Z. Zhang, P. P. Jung, V. Grouès, P. May, C. Linster, E. Glaab, Web-based QTL linkage analysis and bulk segregant analysis of yeast sequencing data, GigaScience (2019), 8(6), 1-18
12. S. Köglsberger, M. L. Cordero-Maldonado, P. Antony, J. I. Forster, P. Garcia, M. Buttini, A. Crawford, E. Glaab, Gender-specific expression of ubiquitin-specific
peptidase 9 modulates tau expression and phosphorylation: possible implications for tauopathies, Molecular Neurobiology (2017), 54(10), pp. 7979
13. Kleiderman, S., Gutbier, S., Ugur Tufekci, K., Ortega, F., Sá, J. V., Teixeira, A. P., et al. (2016). Conversion of Nonproliferating Astrocytes into Neurogenic Neural Stem Cells: Control by FGF2 and Interferon-γ. Stem Cells, 34(12), 2861-2874.
14. Bolognin, S., Fossépré, M., Qing, X., Jarazo, J., Ščančar, J., Moreno, E. L., et al. (2019). 3D Cultures of Parkinson's Disease-Specific Dopaminergic Neurons for High Content Phenotyping and Drug Testing. Advanced Science, 6(1), 1800927.
15. Jaeger, C., Glaab, E., Michelucci, A., Binz, T. M., Koeglsberger, S., Garcia, P., ... & Buttini, M. (2015). The mouse brain metabolome: region-specific signatures and response to excitotoxic neuronal injury. The American Journal of Pathology, 185(6), 1699-1712.
